JOURNAL OF PROPULSION AND POWER Vol. 22, No.

6, NovemberDecember 2006

Review of Multiscale Modeling of Detonation

Joseph M. Powers University of Notre Dame, Notre Dame, Indiana 46556-5637
Issues associated with modeling the multiscale nature of detonation are reviewed, and potential applications to detonation-driven propulsion systems are discussed. It is suggested that a failure of most existing computations to simultaneously capture the intrinsic microscales of the underlying continuum model along with engineering macroscales could in part explain existing discrepancies between numerical predictions and experimental observation. Mathematical and computational strategies for addressing general problems in multiscale physics are rst examined, followed by focus on their application to detonation modeling. Results are given for a simple detonation with one-step kinetics, which undergoes a period-doubling transition to chaos; as activation energy is increased, such a system exhibits larger scales than are commonly considered. In contrast, for systems with detailed kinetics, scales ner than are commonly considered are revealed to be present in models typically used for detonation propulsion systems. Some modern computational strategies that have been recently applied to more efciently capture the multiscale physics of detonation are discussed: intrinsic low-dimensional manifolds for rational ltering of fast chemistry modes, and a wavelet adaptive multilevel representation to lter small-amplitude ne-scale spatial modes. An example that shows the common strategy of relying upon numerical viscosity to lter ne-scale physics induces nonphysical structures downstream of a detonation is given.

Nomenclature
b D D d E f g h J L M N P RAM t t X x x x Y y A = = = = = = = = = = = = = = = = = = = = = = = = = = = = generic constant vector diagonal matrix dimensionless detonation wave speed number of dimensions dimensionless activation energy generic convective and diffusive ux generic reaction source generic function for linear representation generic Jacobian matrix generic length, m local reaction length scale, m number of dependent variables number of spatial discretization points matrix of eigenvectors size of random access memory, byte time, s dimensionless time mole fraction distance coordinate, m generic distance vector dimensionless distance coordinate mass fraction dimensionless distance coordinate eigenfunction amplitude diagonal matrix of eigenvalues eigenvalue, 1/s generic dependent variable generic orthonormal basis function

Subscripts

i j max min

= = = =

eigenfunction counter, species counter spatial discretization counter maximum minimum

Superscript

= constant state

I.

Introduction

VER the past several years, revolutionary advances in computational modeling of multiscale physics13 have prompted numerous numerical simulations of the subject of this special issue, detonation-driven propulsion devices.462 Among the devices in which detonation physics plays a leading role are the ram accelerator, the hypothesized oblique detonation wave engine, and the pulse detonation engine. And whenever shocks are present in the vicinity of chemical reaction, as is the case for many rocket and airbreathing propulsion applications, detonation physics is relevant. The computational advances are embodied in improved mathematical models, numerical solution algorithms, and computational hardware. Most critically, the improvements have enabled a more robust scientic design procedure in which the engineer can predict with greater precision the behavior of a device a priori. The key to past and future enhancements in prognostic ability lies in the increase in the engineers capacity to describe physical events that evolve over a wide range of spatial and temporal scales. While in the popular imagination, the so-called Moores law63 effect, which

Joseph M. Powers received his BSME, MSME, and PhD in 1983, 1985, and 1988, respectively, all in mechanical engineering, from the University of Illinois at Urbana-Champaign. Since 1989 he has been on the faculty at the University of Notre Dame, where he considers problems in reactive uid mechanics, high-speed ows, propulsion, energetic granular solids, numerical methods, and applied mathematics. He has held summer appointments at NASA Glenn Research Center, U.S. Air Force Wright Laboratory, Los Alamos National Laboratory, and Argonne National Laboratory. He is an Associate Fellow of the AIAA, Associate Editor of the Journal of Propulsion and Power, and a long time member of the Propellants and Combustion Technical Committee. He is also a member of SIAM, APS, ASME, and the Combustion Institute, as well as winner of the University of Notre Dame College of Engineering Teaching Award.

Received 26 May 2005; revision received 11 October 2005; accepted for publication 11 December 2005. Copyright c 2006 by Joseph M. Powers. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission. Copies of this paper may be made for personal or internal use, on condition that the copier pay the $10.00 per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923; include the code 0748-4658/06 $10.00 in correspondence with the CCC. 1217

1218

POWERS

describes the exponentially increasing speed of computational hardware, is celebrated as the chief enabler of these enhancements, although in actuality, a quieter, but no less important set of improvements in computational modeling techniques deserves as much credit. These gains range from more efcient compilers, new algorithms to better exploit massively parallel hardware architectures, improved fundamental algorithms for solution of large linear algebraic systems, advances in grid-generation methods, and improved algorithms for solving large sets of ordinary and partial differential equations.1,2 Although the gains of the past decades have been remarkable, this paper will demonstrate that in many respects certain problems in detonation propulsion have barely been dented; many more scales remain to be captured before one can claim true understanding. As will be shown, key physical phenomena involving unsteady and multidimensional effects, such as ignition, stability, transverse wave dynamics, detonation diffraction, detailed reaction zone structures, and diffusive structures, remain unresolved for most important engineering applications. As a result, it is argued that the aeropropulsion modeling community stands to benet from more critical and skeptical examination of reactive ow calculations used in design than is employed at present. It will be recommended, then, that the community continue to nurture advances in algorithms and hardware, but to moderate its expectations so that they are consistent with the scientic limitations of both. In short, the physics are sufciently complex that accurate numerical prediction remains problematic. For example, a recent exercise31 in which several computational algorithms were employed in an effort to reproduce results of a benchmark experiment of a ram accelerator produced, according to the authors, widely different outcomes, with strong sensitivity to induction zone dynamics. Another recent study,49 which included both computational predictions and observations of pulse detonation engines, concluded the important run-up distance to detonation to be underpredicted and that simulation of the deagration to detonation transition (DDT) process remains a challenge. Analogous difculties in related elds are easily found. For example, Kadanoff3 summarizes some notable recent failures in simulations of sonoluminescence, RayleighTaylor instability, and wave breaking. He concludes resolution matters. It is speculated here that challenges associated with truly capturing the multiscale physics in such ows can provide an explanation as to why predictions do not always match data. This paper will review modeling strategies to capture multiscale physics, with attention given to problems that arise in shock-laden supersonic reactive ow such as exists in detonation-driven propulsion devices. It will not give a detailed discussion of individual studies of those devices, which have been discussed in other review articles,5,8,28,30,41,48 or fundamentals of detonation theory, well covered, for example, by Fickett and Davis.64 Nor is it intended for those already expert in multiscale modeling. Instead, it is offered as a summary for those of the general propulsion community who might want to know more about this topic and appreciate how the disciplines of reactive gas dynamics, mathematics, computer science, and aerospace engineering can come together with the goal of advancing propulsion technology. The plan of this paper is to open with a brief review of how multiscale physics is endemic in nature, and in particular, in supersonic propulsion devices. A short discussion then follows of fundamental mathematical modeling strategies that can be used to capture the effects of the physical mechanisms which dominate at broadly different scales. An argument is made that continuum-based mathematical models in the form of partial differential equations will remain for some time the critical tool to describe the behavior of common propulsion devices. A description of popular computational strategies for modeling partial differential equations is then given, focusing on how the multiscale physics is manifested mathematically. The following section gives multiscale detonation modeling examples from previous work of the author and some of his colleagues describing how 1) a simple model of detonation can undergo a classical period-doubling process and transition to a chaotic state,65

2) detailed gas-phase kinetic models give rise to ne-scale structures orders of magnitude below those commonly used in present design analyses,66 3) a detailed gas-phase reaction kinetics model can be rationally reduced so as to lter all ne timescale phenomena below a given threshold,67,68 4) an adaptive-mesh-renement (AMR) technique can be used to simultaneously capture ne and coarse structures in a viscous detonation with detailed kinetics,67 and 5) the effects of articial numerical viscosity can corrupt some postshock detonation oweld structures predicted by models with simple kinetic schemes.69 The paper is closed with some brief comments on engineering applications followed by a short set of conclusions.

II.

General Issues and Strategies in Multiscale Modeling

Here, we give a general discussion of multiscale modeling; a more detailed complementary review is given by Oran and Boris.2
A. Multiscale Physics

Our universe has identiable structures that span a tremendous breadth of length scales. The popular book of Morrison et al.,70 which vividly illustrates the multiscale universe, describes scales ranging from a proton-based scale of 1016 m to a universe diameter of 1025 m. Even more dramatic, physicists are now actively studying a hypothesized quantum gravitational scale of 1035 m (Ref. 71). Moreover, observable phenomena are known to evolve on a broad range of timescales, ranging from 1025 s for the time necessary for a photon to traverse a proton-based length, to 1017 s for the estimated age of the universe.72 An overriding goal of much of science is to develop the most efcient explanation for the behavior of the universe, as well as its components, which spans as wide a set of scales as is practical. The whole being a consequence of the parts, it stands to reason that one path to understanding macrobehavior of the universe is by understanding its building blocks and how they interact. Generally, the wider breadth of scales, the more impractical this strategy becomes.
B. Mathematical Strategies

As the naive theoretical approach of simply describing phenomena at all scales holds little promise, a key to making progress in developing predictive science has been the wise employment of assumptions that decouple phenomena which evolve on one scale from those that evolve on others. The most direct approach to this end combines good intuition guided by experiment with mathematical analysis to either eliminate or rationally account for phenomena that evolve on dramatically different scales than those of the phenomena of interest. Science being ultimately empirical, the proof of the wisdom of such assumptions lies in comparing the predictions of such theories to experiments whose results were not known before the fact. A famous example of such an approach is found in the development of the kinetic theory of gases in the late 19th century. This theory showed how systematic averaging of complex molecular collision processes gives rise to a set of progressively simpler theories, each of which captures fewer physical phenomena, but are often more than sufcient for the task at hand. In a series of analytical steps in which progressively more and more ne scales are removed via averaging, one proceeds from a Liouville to a Boltzmann to a Burnett to a NavierStokes model of reactive uid mechanics.7376 This theory relates both equation of state and diffusive transport properties, valid at the macroscale, to the more fundamental theory based on microscale molecular collisions. For simple molecules, rst principles estimates are available for the resulting constitutive equations that are really models for capturing the average effect of subcontinuum phenomena. The success of this approach has sparked much research in the same analytic vein. One need only consult the recent literature77,78 or any issue of the new journal Multiscale Modeling and Simulation to see that one common conception of multiscale physics is often closely linked to developing custom models for highly distinct

POWERS

1219

physical problems in which multiscale coupling plays a large role. For many problems, this method of the modeling artisan is a useful approach. For others, the basic mathematical models are reasonably well established; however, their solution for general problems is not. Such problems demand the tools of the algorithm artisan to meet the challenges of multiscale physics, which are already intrinsic to the models. The key problems in aeropropulsion tend to recommend the latter approach, as most models for reacting gas dynamics are generally accepted. The main challenges intrinsic in these models arise from their highly nonlinear character, which admits complex multiscale solutions in which disparate scales are often strongly coupled. When one connes consideration to continuum reactive uid mechanics applicable to aeropropulsion devices, one nevertheless nds widespread analytical efforts to lter the scales that remain in the continuum. An estimate of the smallest length scale that is inherent in the resulting continuum models is related to the mean freepath distance between molecular collisions, typically on the order of 107 m. This minimum length scale of relevance is the foundation of length scales predicted by continuum models of mass, momentum, and energy diffusion, as well as elementary reaction kinetics.75 Note further that for compressible ow, the so-called Kolmogorov scale of turbulence is one and the same with the scale of shock thicknesses, which itself is on the order of the mean free-path distance, as can be shown by simple scaling arguments.79 Thus, for a compressible ow with elementary reactions, all nest scales are related to the mean free path. The common amelet approach in which some nescale ame structures are not resolved in some low-Mach-number simulations of combustion loses its physical rationale for a direct numerical simulation80 (DNS) of detonation. In contrast, the largest length scale is typically a small multiple of the device length, which can be 101 m. Within the span of these scales, Fig. 1 gives rough estimates of scales associated with common features in reactive uid mechanics, such as boundary layers and coarse scale reaction zones. This span of eight orders of magnitude of length scale still represents a grand challenge to achieve quantitative predictions from any theory. The list of mathematical strategies to reduce the span of these scales is long and includes methods for ltering both space and

timescales: 1) employment of Euler equations so as to neglect all diffusion processes and their associated thin layersnotably, such theories then admit true shock and contact discontinuities which give rise to new problems when accurate numerical simulation is attempted; 2) use of formal averaging theories to encapsulate nescale behavior in a new constitutive modelsuch approaches are common in ows involving granules and droplets; 3) a wide variety of asymptotic techniques such as boundary layer theory or the method of multiple scales81,82 ; 4) analytic models for Kolmogorovscale turbulent uid interactions; and 5) intrinsic low-dimensional manifolds (ILDM)67,83,84 to locally equilibrate fast chemical reaction modes. Whether or not any further mathematical ltering strategy is taken, the modeler is typically faced with a system of equations, which will be taken to be of the form + f () = g() t (1)

Here, is a generic set of M dependent variables, f is a nonlinear function embodying convective and diffusive uxes, and g is a nonlinear reaction source term. These equations are solved on a spatial domain of x Rd , where d = 1, 2, 3 gives the dimension of the system and x is the spatial variable. Much insight into general multiscale behavior can be gained by considering the special case in which Eq. (1) is linear. In such a case, one can dene an innite set of basis functions, i (x), i = 1, . . . , , which in this linear case are eigenfunctions of the spatial derivative operator. For the linear system, one can always transform to the wave-attached frame so as to remove the convection and be left only with diffusion. Diffusion operators are typically self-adjoint and thus have an innite set of real eigenvalues and orthonormal eigenfunctions. These eigenfunctions span the space, and thus any arbitrary set of initial conditions can be cast as an innite series in terms of the eigenfunctions. The eigenfunctions are typically of a spatially oscillatory nature, whose wave number rises with increasing value of i. One then assumes that the dependent variables can be separated into spatial and temporal modes via the eigenfunction expansion

=
i =1

i (t) i (x)

(2)

Here i (t) is the time-dependent amplitude of the ith eigenmode. Then, Eq. (1) can be cast as an innite set of linear ordinary differential equations (ODEs) of the form di = i i , dt i = 1, . . . , (3)

Fig. 1 Estimate of range of length scales for continuum model of compressible reactive ow with detailed kinetics in typical propulsion devices along with range of scales that can be loaded into desktop computer with 109 bytes RAM for one-, two-, and three-dimensional calculations with M = 15 variables per cell.

which describes the temporal evolution of the amplitude of each eigenmode. Here i is the eigenvalue of the ith eigenmode. Critically, one notes that each ODE in Eq. (3) is uncoupled from all others and thus can be solved individually. Consequently, an analytic solution can be obtained for the amplitude of each mode; one typically nds, given that an entropy inequality is satised, that the amplitude of a given high-frequency mode decays to zero as the mode number increases to innity. Thus, a nite series can accurately approximate the solution, and one can estimate the error incurred by the truncation. Such solutions formally have a multiscale character, but there is no coupling across scales, and the error of neglecting ne scales is small and quantiable. There is no robustly accurate way to segregate the effects of reaction and diffusion. Even in a purely linear version of Eq. (1), the eigenvalues i are consequences of the combined effects of reaction and diffusion. An analysis of the related, operator-split, spatially homogeneous version of Eq. (1), d/dt = g(), when g happens to be a linear operator, gives rise to a nite set of M eigenvalues, whose reciprocals give estimates of a nite set of reaction timescales. These scales can indeed be widely disparate and indicate that chemically induced stiffness is present. The stiffness however is nite and will

1220

POWERS

always be dominated by stiffness induced by the continuous differential diffusion operator when spatial inhomogeneity is included. In fact, only when enough modes are included so that limi i is dominated by diffusion, and so transcends the largest eigenvalue predicted by the spatially homogeneous model, are enough modes present for the solutions series to be in the asymptotically convergent region, so that one can claim the essence of the solution is spatially resolved. A discussion of the dangers of the common practice of splitting chemistry from convection and diffusion is given by LeVeque and Yee.85 In fact, the artice of operator splitting creates the illusion that chemical stiffness can be segregated from that of diffusion, when in fact the two are often highly coupled. If Eq. (1) is nonlinear, one can still perform an expansion of in terms of an innite set of orthonormal basis functions i , which no longer are eigenfunctions of the spatial operator. In this case one can, in principle, approximate the system by an innite set of ODEs of the form di = hi (1 , . . . , ), dt i = 1, . . . , (4)

In the nonlinear case, there is coupling between modes, and it is difcult to determine a priori how strong that coupling is. In practice, it is often nearly impossible to form the exact representation given in Eq. (4), and one usually has to resort to discrete approximations. However, many spectral and pseudospectral methods, as well as loworder nonlinear stability analyses, do take as their starting points truncated versions of Eq. (4).
C. Computational Strategies

For nonlinear problems, exact solutions are typically not available, and the modeler most often turns to discretization of the underlying partial differential equations to enable an approximate numerical solution to be obtained. These equations contain their own challenges for which a summary of approaches will be summarized here. As the detonation-based aeropropulsion literature is beginning to contain more calculations based on NavierStokes models,35,46,61,8691 attention will be focused on them; similar arguments could be made for related models. The ongoing development of the digital computer along with appropriate algorithms over the past decades has enabled a widening of the breadth of scales that can be modeled. However, it is often not appreciated still how narrow the band is at present. Consider a calculation to be performed in a one-, two-, and three-dimensional linear, square, and cubical box. Take d = 1, 2, 3, to be the dimension of the box and L max to be its length. Take the minimum physical length scale to be L min and M to be the number of dependent variables to be calculated at each computational node. Then it is easy to show the necessary size of the random access memory RAM is given by RAM = M L max L min
d

predictions are easier defended, is to always capture the nest scales. In this approach then, the largest one-, two-, and three-dimensional aeropropulsion problems, which can be loaded onto a 109 bytes computer, have characteristic lengths of 101 m, 103 m, and 105 m, respectively. This conservative approach is rarely adopted because it cannot capture the macroscopic scales of interest in typical engineering applications. In the more common, as well as more risky, approach, the engineering scale is taken as the largest scale, and as ne a discretization as is practical is used for a simulation. Assuming the largest device geometric scale is 101 m, a modern desktop computer with 109 bytes RAM can at best capture scales down to 107 m, 103 m, and 101 m in a given one-, two-, and three-dimensional calculation. Phenomena that evolve below these scales are not captured in any detail and are typically overwhelmed by nonphysical numerical diffusion. Considering now supercomputers, recently developed massively parallel systems have achieved as much as 3.2 1013 bytes RAM.92 Even then, the ability to span scales is not what is required for a rst principles scientic design. For such a machine, the ratios of scales that can be spanned are 2 1012 , 1 106 , and 1 104 for one-, two-, and three-dimensional calculations, respectively. The resolution of the necessary ratio of scales of L max /L min = 108 requires 1.5 109 bytes RAM, 1.5 1017 bytes RAM, and 1.5 1025 bytes RAM for a one-, two-, and three-dimensional calculation, respectively. Moreover, these estimates only speak to the necessary ability to load the problem into RAM but ignore the critical issue of speed of the actual computation, which increases dramatically with the neness of the discretization. An indication of the the range of scales that can be resolved by such a DNS as a function of available RAM is given in Fig. 2. With some notion of the range of length scales that can or should be modeled, the numerical analyst typically performs some class of spatial discretization of Eq. (1) in which the solution is only calculated at N discrete points. This allows the spatial differential operators to be replaced by discrete counterparts and transforms the coupled set of partial differential equations into a large set of differential algebraic equations (DAEs) of the form D d j dt = j [ f (1 , . . . , N )] + g( j ) j = 1, . . . , N (6)

(5)

Here j is a discrete counterpart to the continuous gradient operator, and D is a diagonal matrix whose diagonal is composed of zeros or ones. It arises because some methods give rise to purely algebraic relations at some points. If A = I, the identity matrix, as it is for several methods, the set of N DAEs becomes a set of N ODEs. There is a wide choice of spatial discretizations in common use. Among them are 1) nite difference/nite volume, 2) nite element, 3) spectral, 4) wavelet, and 5) manifold. There is a large body of literature on the rst three discretizations, nicely summarized by Iserles93 ; for the hyperbolic equations used commonly for inviscid detonation simulations, LeVeque94 gives a modern discussion. The

assuming a straightforward spatially uniform discretization of the geometric domain. The most desirable calculation is one that has sufcient resolution to capture all of the relevant scales admitted by an underlying mathematical model. Such a calculation is a DNS. The real challenge associated with performing a DNS in a continuum calculation of ow in an aeropropulsion device, even on modern computational hardware, is evident on examination of Fig. 1. This gure presumes that M = 15 variables are associated with each cell, which is a reasonable estimate for calculation with detailed kinetics of, for example, a hydrogen-oxygen system. Estimating that a good desktop computer in the present day can have 109 bytes of RAM, it is seen that such a machine can perform simulations that span a ratio of scales L max /L min = 108 , 104 , and 102 for one-, two-, and three-dimensional calculations, respectively. Only the one-dimensional calculation approaches the breadth of physical scales admitted by a continuum model for aeropropulsion engineering devices. Figure 1 also shows the results of two distinct approaches to modeling. The purpose of the conservative DNS approach, whose

Fig. 2 Estimate of breadth of length scales that can be modeled for a typical continuum calculation of compressible reactive ow with detailed kinetics as a function of available random access memory (RAM).

POWERS

1221

rst two classes are related in that both involve a discretization in which a small number of points in the neighborhood of point j are used to estimate the spatial derivatives. Such discretizations have a low order of spatial accuracy. Spectral methods95,96 involve discretizations in which all points in the domain are used to estimate local values of spatial derivatives; at the expense of additional complexity, these have a high order of accuracy for smooth problems. The wavelet discretization is a relatively new class whose use admits several advantages,67 especially in developing computational grids that dynamically adapt to evolving ow structures. The manifold discretization, described for reactive systems recently,68 is another relatively new approach, which gives rise to a set of DAEs. The algebraic relations force the solution to lie on specied manifolds and amount to equilibrating fast timescale events. Considering from this point on the case where D = I, Eq. (6) is a large set of nonlinear ODEs in time of the form d j dt = j [ f (1 , . . . , N )] + g( j ) j = 1, . . . , N (7)

Here J is the locally constant Jacobian matrix, and b is a locally constant vector. Standard linear systems analysis98 shows that each local eigenmode of Eq. (8) evolves at a rate dictated by the eigenvalues of J and that local solutions take the form j (t) = + {P exp[(t t )] P1 I} J1 b j (9)

Spatial variables do not explicitly appear in Eq. (7). Consequently, one can apply notions from the well-developed theory of nonlinear dynamic systems97 to understand behavior of the spatially discretized system. In particular, Eq. (7) for N 3 contains the essential general property of nonlinearity so that one might expect to achieve a chaotic solution in time. In fact, such behavior is shown explicitly in detonation systems in a recent study.65 Such solutions formally admit an innite number of timescales, and none can be neglected. If a nonlinear system becomes chaotic, then small-scale events, which could safely be neglected if the system were linear, can no longer be simply dismissed. This is the fundamental conundrum of chaos: for such systems, even low-amplitude disturbances at ne scales have the potential to cascade up to induce large disturbances at coarse scales. In such cases, similar to homogeneous turbulence, one must perform more specialized studies to determine what meaningful statistically averaged information can be extracted from the nonlinear system and not expect to recover deterministic predictions at all scales. Even if the solution to the ODEs of Eq. (7) were known to innite precision, the solution might or might not represent well the solution of the corresponding continuous system, Eq. (1). How well the solution to Eq. (1) is approximated depends upon whether the spatial discretization was chosen to be ne enough to capture the intrinsic spatial scales of Eq. (1). This issue can only be resolved in general by systematic study of if and how the approximate solution is converging as the spatial discretization is rened. One can usually be condent in a solution if it is converging at a rate consistent with the order of accuracy of the spatial discretization; however, in the absence of an exact solution one cannot have absolute certainty that one is converging to a correct, physically meaningful solution. Imagine, as a counterexample, a sensitive nonlinear system with multiple stable equilibria. Some equilibria can be physical and some nonphysical. A small error at an intermediate stage of the computation can shift the solution path in phase space and send the solution to an incorrect, and perhaps nonphysical, albeit stable, equilibrium point. Fortunately, experience shows convergence to an incorrect solution is rare for models that are mathematically well posed and whose intrinsic scales have been computationally resolved. Lastly, one notes that 1) a solution is never converged, but only can be converging at an appropriate rate; and 2) only when a solution is converging appropriately is it proper to ask the even more important question for precision engineering of just how computationally accurate is that solution. Important information regarding the nonlinear temporal behavior of a spatially discretized system can often be gleaned from a local linearization, which when applied to Eq. (7) at a generic time and state t , gives rise to the linear system of ODEs of the form j d = J j + b, j j dt j j (t ) = j j = 1, . . . , N (8)

Here P is the matrix whose columns are populated by the right eigenvectors of J, and is the diagonal matrix whose diagonal is composed of the eigenvalues j , j = 1, . . . , N , of J. This solution presumes that J can be diagonalized; if this is not possible, a related Jordan decomposition can be performed. The reciprocal of the real part of each eigenvalue j gives the local timescale of evolution of each local eigenmode. Importantly, it can be shown that increasing the spatial resolution, thus increasing N , has the effect of introducing larger eigenvalues so as to increase their disparity, and thus the stiffness. Consequently, there are more timescales to resolve in a numerical solution. It is Eq. (7) that is the departure point for a wide array of multiscale methods found in the scientic computing literature. It is well known that so-called time-explicit and time-implicit methods can be used to solve such systems. Both require a second discretization, here in time, with explicit methods using known values of j to evaluate the right side of Eq. (7) and implicit methods using unknown values of j to evaluate the same term. Explicit methods can be employed via relatively simple algorithms and do not place excessive demands on storage, but for numerical stability require that the discrete time step be of the order of the reciprocal of the largest eigenvalue of J. For ne grids, this can become prohibitively expensive. Implicit methods require more complicated solution algorithms, typically involving the iterative solution of a set of nonlinear algebraic equations, which places high demands on storage, with the advantage that much larger time steps can be used while retaining numerical stability. Most solution methods for nonlinear algebraic systems require repeated inversion of large matrices. The matrices that arise from stiff sets of ODEs have a large condition number, rendering their inversion computationally difcult. The matrices that arise from nite difference, nite element, or wavelet discretizations are usually sparse, which renders their inversion somewhat easier. In fact, the inversion of such sparse systems poses the critical computational multiscale challenge for a wide part of the scientic computing community. A large body of literature has arisen as a result.99102 Some of the key techniques are so-called Jacobi and successive-overrelaxation iterative techniques, Krylov subspace methods, and multigrid methods. As these methods often require iterative techniques that need to be truncated at some point, it is important to distinguish convergence of an intermediate iterative solver from convergence of the solution itself. For an approximate solution via a time-implicit technique to Eq. (1) to be converged, one must indeed have convergence of the iterative solver at each time step; in addition, one must use sufcient spatial resolution to insure the approximation to itself is converging. Lastly, AMR methods must be mentioned. The literature on this is rapidly developing103 ; moreover, it is beginning to impact the combustion and detonation community.67,8890,104108 The promise of these methods is that computational grids can be judiciously redistributed in such a fashion to maintain a small error of approximation. Precisely how to distinguish just what constitutes an error can pose challenges to an AMR strategy when no a priori knowledge of the exact solution exists. The idea is motivated by the fact that many ows have large regions of small variation, and, consequently, coarse discretizations can be tolerated in these regions. Likewise, in regions of steep gradients ne discretizations are necessary to capture the physics correctly. Unsteady detonation problems are challenging in that a dynamic adaptation is required in which the user cannot have a priori knowledge as to where to adapt. Thus, algorithms must be designed to automatically make decisions as to where to concentrate the discretization. This can pose serious challenges if the user wants to operate in a massively parallel environment. For ows with a small number of regions of steep gradients, AMR can be an effective tool to efciently capture multiscale phenomena, with a

1222

POWERS

potential gain of orders of magnitude in computational time relative to calculations performed on a uniformly ne grid. However, it is easy to imagine that in a complicated multidimensional detonation oweld during which multiple reections occur that the spatial domain could rapidly become saturated with interacting wavefronts, each of which would need to be resolved. Thus, for such problems, any AMR strategy would soon evolve into something close to a uniformly ne grid.

III.

Sample Multiscale Detonation Results

This section will provide a compact review of some of the authors and coworkers work in detonations that have a multiscale character. There is, of course, a wider literature on this topic, much of which is reviewed in the original source material.
A. Transition to Chaos with Simple Kinetics Fig. 3 Numerically generated bifurcation diagram 25 < E < 28.4 for one-dimensional inviscid detonation with one-step kinetics, adapted from Ref. 65.

Perhaps the simplest exposition of the multiscale character of detonations can be demonstrated with a classical model of unsteady one-dimensional detonation of a calorically perfect ideal gas whose irreversible exothermic reaction is described by one-step Arrhenius kinetics.65 There is a long history of studying the linear109,110 and nonlinear111 stability of steady detonations predicted by this simple model; numerical studies of one-dimensional pulsating detonations are also beginning to appear for systems with detailed kinetics, although it is not clear that the nest scales have been resolved.112 Reference 65 considers the unsteady behavior of waves that have a ChapmanJouguet (CJ) character when unsteady terms are neglected. Activation energy is taken as a bifurcation parameter. A standard case is considered in which the dimensionless heat release q = 50 and the ratio of specic heats = 6 . It is difcult to 5 compactly describe the scaling parameters; a description is given in Ref. 65 and references therein. Results of unusually high accuracy are obtained by use of a novel shock-tting scheme coupled with a new fth-order spatial discretization.113 In all cases, 20 points are taken to describe the so-called half-reaction zone length L 1/2 , which is the length in the steady detonation structure at which the reaction progress variable takes on a value of 1 . Shock tting is 2 a viable method because only a single discontinuity ever appears in these ows. Consequently, there is no rst-order corruption of the results as a result of shock-capturing effects of numerical viscosity, and the results have true fth-order accuracy. This highorder accuracy enables the revelation of multiscale behavior in detonations, which is hidden in closely related studies done at lower order.114,115 For low activation energy, the steady detonation is hydrodynamically stable. As activation energy is increased, the detonation becomes unstable. While linear stability theory predicts more and more unstable modes at higher and higher frequencies as activation energy is increased, in this study the only cases considered are those for which the unstable behavior has as its origin a single unstable lowfrequency mode as predicted by the linear stability theory. Within this conned range, interestingly, one nds a very different class of multiscale behavior: the dimensionless detonation wave speed D undergoes a classic period doubling behavior as activation energy is increased. That is to say, more and more low-frequency modes are exhibited. Consequently, to fully capture the behavior of the lowest frequency mode of oscillation, one must integrate for a long time over a large spatial domain; the largest scale admitted becomes ever larger with increasing activation energy, at least in the limited range studied. This period doubling behavior is summarized in the bifurcation diagram of Fig. 3. This gure is qualitatively similar to bifurcation diagrams found in a wide variety of problems in nonlinear dynamics, for example, the famous logistics equation.97 The bifurcation points accumulate with geometrically increasing frequency in the well-documented way of many chaotic systems. The Feigenbaum constant116 for this accumulation rate is 4.66 0.09, in close agreement to the known value of 4.669201. Moreover, for even greater values of activation energy there are windows of relative placidity, in that only a small number of oscillatory modes are present;

again, this is entirely consistent with general theories of nonlinear dynamics. Figure 4 then shows numerical predictions of unsteady detonation wave speed D vs dimensionless time t for a variety of activation energies. For each of the activation energies, a different class of oscillation is predicted, including one that is chaotic.
B. Reaction Zone Length Scales for Steady Inviscid Models with Detailed Kinetics

Turing next to models with detailed kinetics, a recent study66 has considered a standard one-dimensional model of steady inviscid CJ detonation in a hydrogen-air mixture using a model of nine molecular species and 19 elementary reactions. The model was identical in all respects to that used by Shepherd117 as well as Mikolaitis,118 and predictions were completely consistent with their results. The model was posed as a dynamic system of ODEs with the spatial coordinate x as the independent variable. A numerical solution of the ODEs gave a prediction of the reaction zone structure, shown in Fig. 5. In contrast to earlier results, Fig. 5 is plotted on a logarithmic scale, which better reveals the multiscale nature of this seemingly simple system. At small values of distance from the shock in the region known as the induction zone, the minor species are rapidly evolving over small length scales of less than 106 m. It is not surprising that this is similar to the length scale of molecular collisions as the constitutive theory for the detailed kinetics builds on a foundation from collision theory. The global effect of reaction evolves over a broader length scale of around 101 m. An eigenvalue analysis of the local Jacobian matrix, which characterizes the ODEs in space, reveals the precise values of the length scales in the steady detonation. Because the system has a nite number of ODEs, there are a nite number of length scales, each of nite value. These are plotted in Fig. 6. The smallest and largest, around 106 m and 101 m, respectively, predict well the small and largest scales seen in Fig. 5. The implications of this study for unsteady modeling are disturbing. For the researcher to have full condence in the predictions of a model with detailed kinetics, at a minimum a spatial grid resolution below that of the nest length scale must be captured. That is not the case in the bulk of the modern literature, and Ref. 66 lists recent calculations that are underresolved from one to ve orders of magnitude. One must ask why the underresolved studies give results that at least seem plausible. We speculate that it is because combustion systems tend toward stable equilibria. Experience suggests that underresolved calculations approach the same equilibrium state as do resolved calculations, but at a rate dictated by numerical viscosity rather than the underlying mathematical physics model. It seems likely that a model that relies upon numerical viscosity to describe the nest scales runs the risk of improperly describing transient events such as ignition, DDT, and detonation instability. These issues remained to be explored fully. This point is illustrated by the recent work of Fusina et al.,91 which employs a NavierStokes model with detailed kinetics to

POWERS

1223

a)

d)

b)

e)

c)

f)

Fig. 4 Numerically generated detonation velocity D vs using a fth-order discretization coupled with shock tting for one-dimensional inviscid t, detonation with one-step kinetics: a) E = 27.75, period-4; b) E = 27.902, period-6; c) E = 28.035, period-5; d) E = 28.2, period-3; e) E = 28.5, chaotic; and f) E = 28.66, period-3, adapted from Ref. 65.

Fig. 5 Species mole fraction vs distance for steady CJ detonation in inviscid hydrogen-air mixture predicted by a one-dimensional steady model, adapted from Ref. 66.

Fig. 6 Local length scales vs distance for steady CJ detonation in inviscid hydrogen-air mixture predicted by a one-dimensional steady model, adapted from Ref. 66.

study the stability of a two-dimensional ChapmanJouguet oblique detonation wave (ODW). In their study, which is representative of the state of the art of detonation modeling for propulsion systems, the authors take the engineering approach of capturing device-length scales, but are not able to capture the nest viscous and reaction scales. In a detailed grid-resolution study, they conclude The grid resolution . . . is not ne enough to capture all the length scales present, such as the viscous shock thickness, but it is ne enough

for determination of global or macroscopic phenomena such as the ODW angle. Because determination of stability hinges upon proper resolution of length and timescales, it might be too early to conclude that such CJ waves are stable, and the authors are careful to qualify their conclusion on stability accordingly. Indeed, for ow of simpler uid over the same geometry, resolved calculations of Grismer and Powers119 show that a signicant overdrive is necessary to stabilize inviscid oblique detonations predicted with one-step kinetics.

1224 C. Relaxation of Fast Timescale Kinetics with ILDM

POWERS

When a uid particle is both reacting and advecting, the length of the reaction zone is roughly given by the product of the advection velocity and the reaction time. The small reaction zones of the preceding section are direct consequences of fast reactions. One important strategy to reduce the multiscale challenges of reactive uid mechanics is to develop reduced kinetic models. Unfortunately, most of the well-known methods, for instance, those involving socalled steady-state and partial equilibrium assumptions, come with no guarantee that they can robustly capture the results of the detailed kinetic models; that is to say, in such approaches there is no systematic way to adjust parameters so as to converge to the full kinetics solution. Moreover, even after these approaches have been applied, there is no guarantee that fast timescale events have truly been ltered from the system. The relatively new ILDM83 and the related computational singular perturbation84 methods offer a rational approach to systematically reduce detailed kinetics models to simpler systems. The method is built around spatially homogeneous systems and thus focuses on ltering fast timescale events. The method relies on locally linearizing the ODEs that describe the temporal evolution of the reactive system so as to nd a local Jacobian matrix. An eigendecomposition of the Jacobian matrix reveals an ordering of reactions from fastest to slowest along with the directions in composition space associated with fast and slow modes. A slow manifold is then constructed by forcing the manifold to lie orthogonal to all of the eigenvectors associated with the fast modes. The dimension of this manifold is chosen by the modeler. One can then model the behavior of the spatially homogeneous reactive system by projecting from an arbitrary initial condition to a point on the manifold, thus avoiding the small time steps associated with the fast dynamics. One lets the system evolve only on the manifold, where the timescales are relatively slow, and inexpensive explicit methods can be used to calculate the time variation. A small error is introduced in neglecting the fast timescales; if the initial condition is too far from the ILDM, the error can be large, and special care must be taken to avoid this. Figure 7 shows a projection of a composition space for the mass fraction Y of two of the species of a nine-species, 37-step mechanism that describes combustion in H2 O2 Ar systems.67 Here, the equilibrium point, which can be thought of an ILDM of dimension zero, is calculated a priori as is a one-dimensional ILDM. Also shown are projections of several trajectories, each of which rst relax to the one-dimensional ILDM and then to the equilibrium point. On each trajectory an is plotted at equal time intervals. It is seen that these

Fig. 8 Predictions of pressure vs distance at coarse and ne length scales for one-dimensional viscous detonation of H2 O2 Ar with detailed kinetics model, ILDM kinetics reduction, and wavelet adaptive multilevel representation, adapted from Ref. 67.

agglomerate near the ILDM, indicating that the bulk of the time is spent on the ILDM. The same approach can be used to generate ILDMs of higher dimension, which can capture progressively more timescales. An a priori knowledge of the manifold can be employed in a spatiotemporal calculation to reduce computation time at the expense of a small error. This particular ILDM was used in a NavierStokes detonation calculation,67 and the results are shown in Fig. 8, which gives a plot of pressure vs distance at two highly disparate spatial resolutions. In the ne-scale structure plot, the dots indicate the predictions utilizing the ILDM method, and the solid line gives the results of the full kinetic model. At this scale, they are indistinguishable, but there is a small difference. For this calculation, enough points were sufciently close to the ILDM to enable a reduction in computational time of a factor of about two. For systems that were closer to equilibrium throughout the domain, this efciency gain could be much larger.
D. Detonation for Unsteady Viscous Models with Detailed Kinetics

Fig. 7 ILDM projection for a nine-species, 37-step reaction mechanism of spatially homogeneous H2 O2 Ar combustion as a function of YH2 O along with trajectories from full time integration showing relaxation to the manifold and equilibrium. The symbol denotes equally space 0.10-s time intervals. Total time to relax to equilibrium is near 0.1 ms, adapted from Ref. 67.

In addition to describing the ILDM method, Ref. 67 also exposes other multiscale features of detonations. This study considered a NavierStokes model and thus admitted shocks of nite thickness. Here the thickness was a function of the physical viscosity and not the numerical viscosity. The viscous layer associated with the shock is thin, as seen in Fig. 8, which shows in its bottom half a microscale portion of the macroscale given in the top half. The viscous layer actually overlaps with some of the nest reaction zone lengths but is distinct from the better understood induction zone. These calculations were enabled by a relatively new adaptive method known as the wavelet adaptive multilevel representation (WAMR). To briey summarize this complex method, a given set of initial conditions is projected onto a basis of wavelet functions. These functions have localization in both physical and wave-number space. Basis functions whose amplitude is below a threshold value are discarded, except for those in the near neighborhood of the threshold. For systems such as this one that contain a small number of zones with steep gradients, this projection and ltering greatly reduces the number of equations that need be solved, as well as the associated stiffness. The system is then evolved in time. If wavelet

POWERS

1225

amplitudes fall below the threshold value, they are discarded. Those in the near-threshold region whose amplitudes were calculated are checked to see if they cross the threshold, and if so, they are included in the next calculation. As a result of this WAMR method, the entire one-dimensional unsteady viscous detonation with detailed kinetics was represented by no more than 3 102 collocation points at any given time. Had the same calculation been performed on a spatially uniform grid, approximately 107 collocation points would have been required to achieve the same accuracy. Had a similar calculation been performed in multiple dimensions, the efciency of the WAMR method, relative to a uniform mesh, would have been even greater.
E. Two-Dimensional Detonation Structures with Simple Kinetics

A common practice in detonation calculations is to ignore diffusion processes and only consider convection and reaction. In multidimensional detonation studies, such as those of Bourlioux and Majda120 or Williams et al.121 visually striking detonation structures have been predicted with reactive Euler models. Recent studies122,123 have further highlighted the detailed spatiotemporal structures present in seemingly simple phenomena as detonation corner-turning predicted by inviscid one-step models. However, as continuum models of reaction and diffusion both have molecular collision models as their foundation, and both models predict relaxation on the same length scales, it is difcult to physically justify neglecting diffusion without also neglecting reaction. Nevertheless, as it is commonly done, its consequences should be analyzed. One recent simplied analysis was given by Singh et al.69 There, a simple one-step chemistry model for the unsteady twodimensional behavior of a calorically perfect ideal gas nearly identical to that considered by Short and Stewart124 was considered under the conditions found in Ref. 124 to contain one linearly unstable

mode. In contrast to Ref. 124, Ref. 69 considered both viscous and inviscid models. The physical viscosity in the NavierStokes model was adjusted so that the viscous layers were roughly one-tenth the length of the global reaction zone length. Both the Euler and Navier Stokes models were subjected to a grid-renement study, and the results are summarized in Fig. 9. In the Euler calculations, intrinsic numerical viscosity, which depends on the size of the grid and the details of the particular numerical method chosen, always plays a large role in selecting the ow structures that evolve at and downstream of the shock. In the NavierStokes calculations at coarse resolutions, the same articial viscosity dominates the physical viscosity, and the structures depend on the grid. As the grid is rened for the Euler calculations, the articial viscosity decreases, and fewer downstream instabilities, such as the KelvinHelmholtz instability, are suppressed. Thus, it is possible to see ever-ner downstream structures in Fig. 9. At coarse resolutions in the NavierStokes model predicts similar results as the Euler model. In this case, the inherent numerical viscosity of the method dominates the physical viscosity. However, as the grid is rened in the NavierStokes calculations, the physical viscosity comes to dominate the numerical viscosity, and no ner scale structures are apparent. The sensitivity of our results to resolution is in general agreement with the recent related study of Tegn r and Sj green.125 e o Clearly, the downstream structures in detonations are inuenced by the amount of viscosity, real or numerical present. Moreover, these difculties are entirely analogous to those reported by Kadanoff,3 which depend on a numerical viscosity in inviscid calculations of a RayleighTaylor instability; Ref. 3 concludes, The practical meaning is that we cannot promise different approximation approaches will converge to the same answer, and that any one of those will correspond to the experimental system. The most

Fig. 9 Isochores at three different spatial resolutions at three different times, 1 , 2 , 3 , for two-dimensional Euler and NavierStokes detonation with t t t one-step Arrhenius kinetics. Domain for each simulation, [0, 15], [0, 20], adapted from Ref. 69. x y

1226

POWERS

straightforward remedy is to employ physical viscosity and rene the grid so that it dominates over numerical viscosity. However, Ref. 126 has noted that the dynamics of the leading shock, which are those that determine the patterns etched on walls that have long been observed in experiment, seem to be insensitive to the magnitude of articial viscosity in Euler calculations. Although fortuitous, there is no guarantee that this result will extend to other important issues. Also, as noted in Ref. 127, one will in very special cases nd that numerical viscosity captures the effects of physical viscosity. However, with no a priori standard of what a viscous structure should be, it is unlikely that one could discern that grid resolution would lead to the correct calculation, and so this does not appear to be of great practical interest.

that some numerical solutions can be both inaccurate and important. Such a notion, detailed in his article which highlights astrophysical detonation, is not inconsistent with that of Buckmaster,129 who holds for more conventional systems, Modelers write down false equations and extract useful, physically relevant information from them. Nor is it inconsistent with Williams130 trio of maxims: Theory neednt be right to be good...Theory neednt be mathematical to be right...Theory neednt be incomprehensible to be mathematical. None are arguing for inaccurate design calculations relative to the highly desirable veried and validated prediction.131,132 Instead, they remind the reader that if enhanced scientic insight is the goal of the calculation, that such an end can be achieved by making, what are for design purposes, inaccurate calculations.

IV.

Detonation Propulsion Implications

What then are the implications for detonation applications in propulsion? In the most general sense, the answer cannot be known with certainty until one has performed calculations that actually capture the full multiscale nature of the ow, from molecular collision scales to device scales. However, it is possible to offer some speculation. It must be said that many important parameters for system performance, such as steady detonation wave speed and nal detonation pressure, do not have a strong dependence on multiscale dynamics and instead depend mainly on thermodynamic properties. However, whenever transient events are relevant, such as in ignition, a proper capturing of the multiscale space and timescales is critical to predict device performance. Consider, for example, the pulse detonation engine.41 This engine relies upon the successful initiation of a detonation at a rate of many cycles per second. In developing models for such devices, an understanding of the ignition and DDT behavior is critical. Models of actual reactive mixtures have heretofore not had great success in this prediction.49 One cannot yet rule out a failure to capture the multiscale nature of the ignition/DDT process as a reason for this lack of success. In short, one must resolve the local timescales for chemical power deposition and radical generation relative to the local acoustic timescales in the spatial domain in which the transient events are evolving. One might expect inertial connement for fast timescale events and wave generation for slower events. Many recent studies of pulse detonation engines have focused on the use of small geometric obstacles, such as Schelkin spirals, to enhance the DDT process. Modeling of how shocks and detonations diffract around such small-scale barriers so as to induce ignition is a process in which multiscale physics clearly play a large role. Successful transmission of a detonation from a large tube to a small tube without inducing extinction is an issue in pulse detonation engines. And once again, a proper numerical capture of this highly transient multidimensional effect represents a challenge for multiscale methods. Another issue relating to the pulse detonation engine is the issue of cellular detonation instability and transverse waves. Although such waves can be predicted with models of one-step kinetics, much uncertainty remains with regard to models of detailed kinetics to be able to match the well-known results of fundamental experiments. It is again likely that the present-day inability to capture ne-scale phenomena lies at the heart of our inability to correctly predict these structures. Certainly, one can say that grid-renement studies in onestep kinetic models reveal that the structures are sensitive to a proper resolution of the reaction zones, and the same is likely true for models with detailed kinetics. As a pulse detonation engine relies for thrust on the axial transmission of impulse, any impulse diverted to transverse modes is that which cannot be used to generate thrust, and so any theory that can suggest how that impulse can be directed in a useful way will have value. Similar comments could be made with regard to the ram accelerator, the oblique detonation wave engine, and other devices. In any device in that detonations must be initiated, where they can diffract, where the detonation becomes unstable, or where extinction could exist, an understanding of the fundamental science of multiscale detonations has direct ramications for device performance. All this said, it must be remembered that not all calculations are done for engineering design, and, as subtly argued by Kadanoff,128

V.

Conclusions

It is clear that observable detonation physics is richly complicated with spatial and temporal structures evolving on a wide range of scales. It is also clear that the nonlinear continuum theories that are commonly used to mathematically model these physics have the ability to predict many of the basic observable quantiable features, such as one-dimensional detonation wave speeds, as well as some qualitative details, such as cellular structures. How to explain those differences that remain, in evidence in the widely different outcomes of Ref. 31, is a grand challenge. One plausible hypothesis is that the underlying continuum mathematical models are in fact correct and that it remains for us to devise better ways to truly capture all of the scales these models admit. Such is the nature of the mathematical exercise known as verication.131,132 Only after a model has been veried is it appropriate to ask the deeper scientic question regarding validation: do the equations accurately predict what is observed in experiment? In short, verication considers if the equations were solved correctly, and validation considers if the correct equations were solved. And if a veried model cannot be validated, it is only then suitable to begin to question the modeling assumptions, constitutive theories, and material property values. The practice of harmonizing mathematically unveried calculations with experimental observation by tuning adjustable parameters is too common and often leads to models that are at best postdictive rather than the preferred predictive. Such hasty calculations, when otherwise done carefully, occasionally serve immediate needs, but do little to further the progress of scientic aeropropulsion design, nor do they do much to advance fundamental science. At best they provide a useful tool to interpolate experimental results. At worst, calculations performed with intemperate attention to both verication and validation serve neither science nor engineering and can, in extreme cases, pose a real danger to those who rely on such calculations. To conclude, the science-based engineering design process requires signicant effort, as do most enterprises that seek to truly expand the frontiers of what can be achieved. Indeed, it will take some time and further advances in both hardware and algorithms before rst principles models that are both veried and validated will be able to be used with condence to quantitatively predict, with no a priori knowledge of the outcome, the detailed behavior of detonation-driven propulsion devices. At a minimum for the present, one can hope that the varied communities involved in this enterprise can appreciate the challenges and limitations imposed by the multiscale physics on both theories and computational methods. But it should also be realized that tremendous progress has been made in the past in expanding the spectrum of scales that can be modeled, and there is no reason why we should not continue to seek new ways to expand them even further.

Acknowledgments
The author is grateful for the help of many of his students and colleagues in working on these problems over the years, including Tariq D. Aslam, Keith A. Gonthier, Matthew J. Grismer, Andrew K. Henrick, Samuel Paolucci, Yevgeniy Rastigejev, and Sandeep Singh, as well as the past support of the National Science Foundation, the Air Force Ofce of Scientic Research, NASA, and Los

POWERS

1227

Alamos National Laboratory. The author also thanks the reviewers for several comments that have been incorporated into the text.

References
E. S., Matchsticks, Scramjets, and Black Holes: Numerical Simulation Faces Reality, AIAA Journal, Vol. 40, No. 8, 2002, pp. 14811494. 2 Oran, E. S., and Boris, J. P., Numerical Simulation of Reactive Flow, 2nd ed., Cambridge Univ. Press, Cambridge, England, UK, 2001. 3 Kadanoff, L. P., Excellence in Computer Simulation, Computing in Science and Engineering, Vol. 6, No. 2, 2004, pp. 5767. 4 Cambier, J. L., Adelman, H., and Menees, G. P., Numerical Simulations of an Oblique Detonation-Wave Engine, Journal of Propulsion and Power, Vol. 6, No. 3, 1990, pp. 315323. 5 Eidelman, S., Grossman, W., and Lottati, I., Review of Propulsion Applications and Numerical Simulations of the Pulsed Detonation Engine Concept, Journal of Propulsion and Power, Vol. 6, No. 7, 1991, pp. 857865. 6 Yungster, S., Eberhardt, S., and Bruckner, A. P., Numerical Simulation of Hypervelocity Projectiles in Detonable Gases, AIAA Journal, Vol. 29, No. 2, 1991, pp. 187199. 7 Yungster, S., and Bruckner, A. P., Computational Studies of a Superdetonative Ram Accelerator Mode, Journal of Propulsion and Power, Vol. 8, No. 2, 1992, pp. 457463. 8 Rubins, P. M., and Bauer, R. C., Review of Shock-Induced Supersonic Combustion Research and Hypersonic Applications, Journal of Propulsion and Power, Vol. 10, No. 5, 1994, pp. 593601. 9 Buckmaster, J. D., Jackson, T. L., and Kumar, A., Combustion in HighSpeed Flows, Kluwer, Dordrecht, The Netherlands, 1994. 10 Grismer, M. J., and Powers, J. M., Calculations for Steady Propagation of a Generic Ram Accelerator Conguration, Journal of Propulsion and Power, Vol. 11, No. 1, 1995, pp. 105111. 11 Li, C., Kailasanath, K., Oran, E. S., Landsberg, A. M., and Boris, J. P., Dynamics of Oblique Detonations in Ram Accelerators, Shock Waves, Vol. 5, Nos. 12, 1995, pp. 97101. 12 Leblanc, J. E., Lefebvre, M. H., and Fujiwara, T., Detailed Flowelds of a RAMAC Device in H2 O2 Full Chemistry, Shock Waves, Vol. 6, No. 2, 1996, pp. 8592. 13 Nusca, M. J., and Kruczynski, D. L., Reacting Flow Simulation for a Large-Scale Ram Accelerator, Journal of Propulsion and Power, Vol. 12, No. 1, 1996, pp. 6169. 14 Ashford, S. A., and Emanuel, G., Oblique Detonation Wave Engine Performance Prediction, Journal of Propulsion and Power, Vol. 12, No. 2, 1996, pp. 322327. 15 Dyne, B. R., and Heinrich, J. C., Finite Element Analysis of the Scramaccelerator with Hydrogen Oxygen Combustion, Journal of Propulsion and Power, Vol. 12, No. 2, 1996, pp. 336340. 16 Ahuja, J. K., Kumar, A., Singh, D. J., and Tiwari, S. N., Simulation of Shock-Induced Combustion past Blunt Projectiles Using Shock-Fitting Technique, Journal of Propulsion and Power, Vol. 12, No. 3, 1996, pp. 518526. 17 Saurel, R., Numerical Analysis of a Ram Accelerator Employing TwoPhase Combustion, Journal of Propulsion and Power, Vol. 12, No. 4, 1996, pp. 708717. 18 Li, C., Kailasanath, K., and Oran, E. S., Stability of Projectiles in Thermally Choked Ram Accelerators, Journal of Propulsion and Power, Vol. 12, No. 4, 1996, pp. 807809. 19 Matsuo, A., and Fujii, K., Detailed Mechanism of the Unsteady Combustion Around Hypersonic Projectiles, AIAA Journal, Vol. 34, No. 10, 1996, pp. 20822089. 20 Li, C., Kailasanath, K., and Oran, E. S., Detonation Structures Generated by Multiple Shocks on Ram-Accelerator Projectiles, Combustion and Flame, Vol. 108, Nos. 12, 1997, pp. 173186. 21 Yungster, S., and Radhakrishnan, K., Computational Study of Reacting Flow Establishment in Expansion Tube Facilities, Shock Waves, Vol. 7, No. 6, 1997, pp. 335342. 22 Yungster, S., Radhakrishnan, K., and Rabinowitz, M. J., Reacting Flow Establishment in Ram Accelerators: A Numerical Study, Journal of Propulsion and Power, Vol. 14, No. 1, 1998, pp. 1017. 23 Choi, J. Y., Jeung, I. S., and Yoon, Y., Numerical Study of Scram Accelerator Starting Characteristics, AIAA Journal, Vol. 36, No. 6, 1998, pp. 10291038. 24 Dudebot, R., Sislian, J. P., and Oppitz, R., Numerical Simulation of Hypersonic Shock-Induced Combustion Ramjets, Journal of Propulsion and Power, Vol. 14, No. 6, 1998, pp. 869879. 25 Choi, J. Y., Jeung, I. S., and Yoon, Y., Unsteady-State Simulation of Model Ram Accelerator in Expansion Tube, AIAA Journal, Vol. 37, No. 5, 1999, pp. 537543. 26 Kailasanath, K., and Patnaik, G., Performance Estimates of Pulsed Detonation Engines, Proceedings of the Combustion Institute, Vol. 28, Pt. 1, 2000, pp. 595601.
1 Oran,

27 Li, C., and Kailasanath, K., Detonation Transmission and Transition in Channels of Different Sizes, Proceedings of the Combustion Institute, Vol. 28, Pt. 1, 2000, pp. 603609. 28 Nettleton, M. A., The Applications of Unsteady, Multi-Dimensional Studies of Detonation Waves to Ram Accelerators, Shock Waves, Vol. 10, No. 1, 2000, pp. 922. 29 Alkam, M. K., and Butler, P. B., Analysis of a Pulsed Detonation Thermal Spray Applicator, Combustion Science and Technology, Vol. 159, 2000, Nos. 16, pp. 1737. 30 Kailasanath, K., Review of Propulsion Applications of Detonation Waves, AIAA Journal, Vol. 38, No. 9, 2000, pp. 16981708. 31 Leblanc, J. E., Nusca, M., Wang, X., Seiler, F., Sugihara, M., and Fujiwara, T., Numerical Simulation of the RAMAC Benchmark Test, Journal de Physique IV, Vol. 10, No. 11, 2000, pp. 119130. 32 Choi, J. Y., Lee, B. J., and Jeung, I. S., Computational Modeling of High Pressure Combustion Mechanism in Scram Accelerator, Journal de Physique IV, Vol. 10, No. 11, 2000, pp. 131141. 33 Moon, G. W., Jeung, I. S., Yoon, Y., Seiler, F., Patz, G., Smeets, G., and Srulijes, J., Numerical Modeling and Simulation of RAMAC30 Experiment Carried Out at the French-German Research Institute of Saint-Louis, Journal de Physique IV, Vol. 10, No. 11, 2000, pp. 143153. 34 Sellam, M., and Forestier, A. J., Pulsed Detonation Engine. Numerical Study, Journal de Physique IV, Vol. 10, No. 11, 2000, pp. 165174. 35 Yungster, S., and Radhakrishnan, K., Simulation of Unsteady Hypersonic Combustion Around Projectiles in an Expansion Tube, Shock Waves, Vol. 11, No. 3, 2001, pp. 167177. 36 Sislian, J. P., Schirmer, H., Dudebout, R., and Schumacher, J., Propulsive Performance of Hypersonic Oblique Detonation Wave and ShockInduced Combustion Ramjets, Journal of Propulsion and Power, Vol. 17, No. 3, 2001, pp. 599604. 37 Nusca, M. J., Numerical Simulation of the Ram Accelerator Using a New Kinetics Mechanism, Journal of Propulsion and Power, Vol. 18, No. 1, 2002, pp. 4452. 38 Ebrahimi, H. B., Mohanraj, R., and Merkle, C. L., Multilevel Analysis of Pulsed Detonation Engines, Journal of Propulsion and Power, Vol. 18, No. 2, 2002, pp. 225232. 39 Ebrahimi, H. B., and Merkle, C. L., Numerical Simulation of a Pulse Detonation Engine with Hydrogen Fuels, Journal of Propulsion and Power, Vol. 18, No. 5, 2002, pp. 10421048. 40 Kailasanath, K., Patnaik, G., and Li, C., The Floweld and Performance of Pulse Detonation Engines, Proceedings of the Combustion Institute, Vol. 29, Pt. 2, 2002, pp. 28552862. 41 Kailasanath, K., Recent Developments in the Research on Pulse Detonation Engines, AIAA Journal, Vol. 41, No. 2, 2003, pp. 145159. 42 Wu, Y. H., Ma, F. H., and Yang, V., System Performance and Thermodynamic Cycle Analysis of Airbreathing Pulse Detonation Engines, Journal of Propulsion and Power, Vol. 19, No. 4, 2003, pp. 556567. 43 He, X., and Karagozian, A. R., Numerical Simulation of Pulse Detonation Engine Phenomena, Journal of Scientic Computing, Vol. 19, Nos. 13, 2003, pp. 201224. 44 Kim, J. H., Yoon, Y., and Jeung, I. S., Numerical Study of Mixing Enhancement by Shock Waves in Model Scramjet Engine, AIAA Journal, Vol. 41, No. 6, 2003, pp. 10741080. 45 Ramadan, K., and Butler, P. B., A Two-Dimensional Axisymmetric Flow Model for the Analysis of Pulsed Detonation Thermal Spraying, Combustion Science and Technology, Vol. 175, No. 9, 2003, pp. 16491677. 46 Kawai, S., and Fujiwara, T., Numerical Analyses of First and Second Cycles of Oxyhydrogen Pulse Detonation Engine, AIAA Journal, Vol. 41, No. 10, 2003, pp. 20132019. 47 Ramadan, K., and Butler, P. B., Analysis of Gas Flow Evolution and Shock Wave Decay in Detonation Thermal Spraying Systems, Journal of Thermal Spray Technology, Vol. 13, No. 2, 2004, pp. 239247. 48 Roy, G. D., Frolov, S. M., Borisov, A. A., and Netzer, D. W., Pulse Detonation Propulsion: Challenges, Current Status, and Future Perspective, Progress in Energy and Combustion Science, Vol. 30, No. 6, 2004, pp. 545672. 49 Tangirala, V. E., Dean, A. J., Chapin, D. M., Pinard, P. F., and Varatharajan, B., Pulsed Detonation Engine Processes: Experiments and Simulations, Combustion Science and Technology, Vol. 176, No. 10, 2004, pp. 17791808. 50 Fan, H. Y., and Lu, F. K., Comparison of Detonation Processes in a Variable Cross Section Chamber and a Simple Tube, Journal of Propulsion and Power, Vol. 21, No. 1, 2005, pp. 6575. 51 Schwartzentruber, T. E., Sislian, J. P., and Parent, B., Suppression of Premature Ignition in the Premixed Inlet Flow of a Shcramjet, Journal of Propulsion and Power, Vol. 21, No. 1, 2005, pp. 8794. 52 Weng, C. S., and Gore, J. P., A Numerical Study of Two- and ThreeDimensional Detonation Dynamics of Pulse Detonation Engine by the CE/SE Method, Acta Mechanica Sinica, Vol. 21, No. 1, 2005, pp. 3239.

1228

POWERS
80 Vervisch, L., and Poinsot, T., Direct Numerical Simulation of NonPremixed Turbulent Flames, Annual Review of Fluid Mechanics, Vol. 30, 1998, pp. 655691. 81 Van Dyke, M., Perturbation Methods in Fluid Mechanics, Parabolic Press, Stanford, CA, 1964. 82 Kevorkian, J. K., and Cole, J. D., Multiple Scale and Singular Perturbation Methods, Springer-Verlag, Berlin, 1996. 83 Maas, U., and Pope, S. B., Simplifying Chemical Kinetics: Intrinsic Low-Dimensional Manifolds in Composition Space, Combustion and Flame, Vol. 88, Nos. 34, 1992, pp. 239264. 84 Lam, S. H., Using CSP to Understand Complex Chemical Kinetics, Combustion Science and Technology, Vol. 89, Nos. 56, 1993, pp. 375404. 85 LeVeque, R. J., and Yee, H. C., A Study of Numerical Methods for Hyperbolic Conservation Laws with Stiff Source Terms, Journal of Computational Physics, Vol. 86, No. 1, 1990, pp. 187210. 86 Clarke, J. F., Kassoy, D. R., and Riley, N., On the Direct Initiation of a Plane Detonation Wave, Proceedings of the Royal Society of London, Series A, Mathematical Physical and Engineering Sciences, Vol. 408, No. 1834, 1986, pp. 129148. 87 Clarke, J. F., Kassoy, D. R., Meharzi, N. E., Riley, N., and Vasantha, R., On the Evolution of Plane Detonations, Proceedings of the Royal Society of London, Series A, Mathematical Physical and Engineering Sciences, Vol. 429, No. 1877, 1990, pp. 259283. 88 Khokhlov, A. M., and Oran, E. S., Numerical Simulation of Detonation Initiation in a Flame Brush: The Role of Hot Spots, Combustion and Flame, Vol. 119, No. 4, 1999, pp. 400416. 89 Oran, E. S., and Khokhlov, A. M., Deagrations, Hot Spots, and the Transition to Detonation, Philosophical Transactions of the Royal Society of London Series A-Mathematical Physical and Engineering Sciences, Vol. 357, No. 1764, 1999, pp. 35393551. 90 Gamezo, V. N., Khokhlov, A. M., and Oran, E. S., The Inuence of Shock Bifurcations on Shock-Flame Interactions and DDT, Combustion and Flame, Vol. 126, No. 4, 2001, pp. 18101826. 91 Fusina, G., Sislian, J. P., and Parent, B., Formation and Stability of Near ChapmanJouguet Standing Oblique Detonation Waves, AIAA Journal, Vol. 43, No. 7, 2005, pp. 15911604. 92 Gara, A., Blumrich, M. A., Chen, D., Ghiu, G. L.-T., Coteus, P., Giampapa, M. E., Haring, R. A., Heidelberger, P., Hoenicke, D., Kopcsay, G. V., Liebsch, T. A., Ohmacht, M., Steinmacher-Burow, B. D., Takken, T., and Vranas, P., Overview of the Blue Gene/L System Architecture, IBM Journal of Research and Development, Vol. 49, Nos. 23, 2005, pp. 195212. 93 Iserles, A., A First Course in the Numerical Analysis of Differential Equations, Cambridge Univ. Press, Cambridge, England, UK, 1996. 94 LeVeque, R. J., Finite Volume Methods for Hyperbolic Problems, Cambridge Univ. Press, Cambridge, England, UK, 2002. 95 Gottlieb, D., and Orszag, S., Numerical Analysis of Spectral Methods: Theory and Applications, Society for Industrial and Applied Mathematics CBMS, Philadelphia, 1977. 96 Fornberg, B., A Practical Guide to Pseudospectral Methods, Cambridge Univ. Press, Cambridge, England, UK, 1998. 97 Drazin, P. G., Nonlinear Systems, Cambridge Univ. Press, Cambridge, England, UK, 1992. 98 Strang, G., Linear Algebra and Is Applications, 3rd ed., Harcourt Brace Jovanovich, Fort Worth, TX, 1988. 99 Golub, G. H., and Van Loan, C. F., Matrix Computations, Johns Hopkins Univ. Press, Baltimore, MD, 1996. 100 Knoll, D. A., McHugh, P. R., and Keyes, D. E., Newton-Krylov Methods for Low-Mach Number Compressible Combustion, AIAA Journal, Vol. 34, No. 5, 1996, pp. 961967. 101 Knoll, D. A., and Keyes, D. E., Jacobian-Free Newton-Krylov Methods: A Survey of Approaches and Application, Journal of Computational Physics, Vol. 193, No. 2, 2004, pp. 357397. 102 Thomas, J. L., Diskin, B., and Brandt, A., Textbook Multigrid Efciency for Fluid Simulations, Annual Review of Fluid Mechanics, Vol. 35, 2003, pp. 317340. 103 Plewa, T., Linde, T. J., and Weirs, V. G., Adaptive Mesh Renement, Theory and Applications: Proceedings of the Chicago Workshop on Adaptive Mesh Renement Methods, Springer-Verlag, Berlin, 2005. 104 Quirk, J. J., A Parallel Adaptive Grid Algorithm for Computational Shock Hydrodynamics, Applied Numerical Mathematics, Vol. 20, No. 4, 1996, pp. 427453. 105 Eckett, C. A., Quirk, J. J., and Shepherd, J. E., The Role of Unsteadiness in Direct Initiation of Gaseous Detonations, Journal of Fluid Mechanics, Vol. 421, 2000, pp. 147183. 106 Becker, R., Braack, M., and Rannacher, R., Numerical Simulation of Laminar Flames at Low Mach Number by Adaptive Finite Elements, Combustion Theory and Modelling, Vol. 3, No. 3, 1999, pp. 503534. 107 Braack, M., and Ern, A., Coupling Multimodelling with Local Mesh Renement for the Numerical Computation of Laminar Flames, Combustion Theory and Modelling, Vol. 8, No. 4, 2004, pp. 771788.

53 Radulescu, M. I., and Hanson, R. K., Effect of Heat Loss on PulseDetonation-Engine Flow Fields and Performance, Journal of Propulsion and Power, Vol. 21, No. 2, 2005, pp. 274285. 54 Perkins, H. D., and Sung, C.-J., Effects of Fuel Distribution on Detonation Tube Performance, Journal of Propulsion and Power, Vol. 21, No. 3, 2005, pp. 539545. 55 Morris, C. I., Numerical Modeling of Single-Pulse Gasdynamics and Performance of Pulse Detonation Rocket Engines, Journal of Propulsion and Power, Vol. 21, No. 3, 2005, pp. 527538. 56 Ma, F. H., Choi, J. Y., and Yang, V., Thrust Chamber Dynamics and Propulsive Performance of Single-Tube Pulse Detonation Engines, Journal of Propulsion and Power, Vol. 21, No. 3, 2005, pp. 512526. 57 Cheatham, S., and Kailasanath, K., Single-Cycle Performance of Idealized Liquid-Fueled Pulse Detonation Engines, AIAA Journal, Vol. 43, No. 6, 2005, pp. 12761283. 58 Owens, Z. C., Mattison, D. W., Barbour, E. A., Morris, C. I., and Hanson, R. K., Floweld Characterization and Simulation Validation of Multiple Geometry PDEs Using Cesium-Based Velocimetry, Proceedings of the Combustion Institute, Vol. 30, Pt. 2, 2005, pp. 27912798. 59 Mattison, D. W., Oehischlaeger, M. A., Morris, C. I., Owens, Z. C., Barbour, E. A., Jeffries, J. B., and Hanson, R. K., Evaluation of Pulse Detonation Engine Modelling Using Laser-Based Temperature and OH Concentration Measurements, Proceedings of the Combustion Institute, Vol. 30, Pt. 2, 2005, pp. 27992807. 60 Tangirala, V. E., Dean, A. J., Pinard, P. F., and Varatharajan, B., Investigations of Cycle Processes in a Pulsed Detonation Engine Operating on Fuel-Air Mixtures, Proceedings of the Combustion Institute, Vol. 30, Pt. 2, 2005, pp. 28172824. 61 Ess, P. R., Sislian, J. P., and Allen, C. B., Blunt-Body Generated Detonation in Viscous Hypersonic Ducted Flows, Journal of Propulsion and Power, Vol. 21, No. 4, 2005, pp. 667680. 62 Ma, F. H., Choi, J. Y., and Yang, V., Thrust Chamber Dynamics and Propulsive Performance of Multitube Pulse Detonation Engines, Journal of Propulsion and Power, Vol. 21, No. 4, 2005, pp. 681691. 63 Moore, G. E., Cramming More Components onto Integrated Circuits, Electronics, Vol. 38, No. 8, 1965, pp. 114117. 64 Fickett, W., and Davis, W. C., Detonation, Univ. of California Press, Berkeley, 1979. 65 Henrick, A. K., Aslam, T. D., and Powers, J. M., Simulations of Pulsating One-Dimensional Detonations with True Fifth Order Accuracy, Journal of Computational Physics, Vol. 213, No. 1, 2006, pp. 311329. 66 Powers, J. M., and Paolucci, S., Accurate Spatial Resolution Estimates for Reactive Supersonic Flow with Detailed Chemistry, AIAA Journal, Vol. 43, No. 5, 2005, pp. 10881099. 67 Singh, S., Rastigejev, Y., Paolucci, S., and Powers, J. M., Viscous Detonation in H2 O2 Ar Using Intrinsic Low-Dimensional Manifolds and Wavelet Adaptive Multilevel Representation, Combustion Theory and Modelling, Vol. 5, No. 2, 2001, pp. 163184. 68 Singh, S., Powers, J. M., and Paolucci, S., On Slow Manifolds of Chemically Reacting Systems, Journal of Chemical Physics, Vol. 117, No. 4, 2002, pp. 14821496. 69 Singh, S., Powers, J. M., and Paolucci, S., Detonation Solutions from Reactive NavierStokes Equations, AIAA Paper 99-0966, 1999. 70 Morrison, P., Morrison, P., Eames, C., and Eames, R., Powers of Ten: About the Relative Size of Things in the Universe, Scientic American Library, New York, 1994. 71 Ball, P., Quantum Gravity: Back to the Future, Nature, Vol. 427, No. 6974, 2004, pp. 482484. 72 Sneden, C., The Age of the Universe, Nature, Vol. 229, No. 6821, 2001, pp. 673675. 73 Hirschfelder, J. O., Curtis, C. F., and Bird, R. B., Molecular Theory of Gases and Liquids, Wiley, New York, 1954, pp. 441667. 74 Grad, H., Principles of the Kinetic Theory of Gases, Handbuch der Physik, edited by S. Fl gge, Vol. 12, Springer-Verlag, Berlin, 1958, u pp. 205294. 75 Vincenti, W. G., and Kruger, C. H., Introduction to Physical Gas Dynamics, Wiley, New York, 1965. 76 Chapman, S., and Cowling, T. G., The Mathematical Theory of NonUniform Gases, 3rd ed., Cambridge Univ. Press, Cambridge, England, UK, 1970. 77 Brandt, A., Bernholc, J., and Binder, K., Multiscale Computational Methods in Chemistry and Physics, Vol. 177, NATO Science Series III: Computer and Systems Sciences, IOS Press, Amsterdam, 2001. 78 Attinger, S., and Koumoutsakos, P., Multiscale Modelling and Simulation, Springer-Verlag, Berlin, 2004. 79 Moin, P., and Mahesh, K., Direct Numerical Simulation: A Tool in Turbulence Research, Annual Review of Fluid Mechanics, Vol. 30, 1998, pp. 539578.

POWERS
108 Sachdev, J. S., Groth, C. P. T., and Gottlieb, J. J., A Parallel SolutionAdaptive Scheme for Multi-Phase Core Flows in Solid Propellant Rocket Motors, International Journal of Computational Fluid Dynamics, Vol. 19, No. 2, 2005, pp. 159177. 109 Erpenbeck, J. J., Stability of Steady-State Equilibrium Detonations, Physics of Fluids, Vol. 5, No. 5, 1962, pp. 604614. 110 Lee, H. I., and Stewart, D. S., Calculations of Linear Detonation Instability: One-Dimensional Instability of Plane Detonation, Journal of Fluid Mechanics, Vol. 216, 1990, pp. 103132. 111 Bourlioux, A., Majda, A. J., and Roytburd, V., Theoretical and Numerical Structure for Unstable One-Dimensional Detonations, SIAM Journal on Applied Mathematics, Vol. 51, No. 2, 1991, pp. 303343. 112 Yungster, S., and Radhakrishnan, K., Pulsating One-Dimensional Detonations in Hydrogen-Air Mixtures, Combustion Theory and Modelling, Vol. 8, No. 4, 2004, pp. 745770. 113 Henrick, A. K., Aslam, T. D., and Powers, J. M., Mapped Weighted Essentially Non-Oscillatory Schemes: Achieving Optimal Order near Critical Points, Journal of Computational Physics, Vol. 207, No. 2, 2005, pp. 542567. 114 Kasimov, A. R., and Stewart, D. S., On the Dynamics of SelfSustained One-Dimensional Detonations: A Numerical Study in the ShockAttached Frame, Physics of Fluids, Vol. 16, No. 10, 2004, pp. 35663578. 115 Ng, H. D., Higgins, A. J., Kiyanda, C. B., Radulescu, M. I., Lee, J. H. S., Bates, K. R., and Nikiforakis, N., Nonlinear Dynamics and Chaos Analysis of One-Dimensional Pulsating Detonations, Combustion Theory and Modelling, Vol. 9, No. 1, 2005, pp. 159170. 116 Feigenbaum, M. J., The Universal Metric Properties of Nonlinear Transformations, Journal of Statistical Physics, Vol. 21, No. 6, 1979, pp. 669706. 117 Shepherd, J. E., Chemical Kinetics of Hydrogen-Air-Diluent Detonations, Dynamics of Explosions, edited by J. R. Bowen, J.-C. Leyer, and R. I. Soloukhin, Vol. 106, Progress in Astronautics and Aeronautics, AIAA, New York, 1986, pp. 263293. 118 Mikolaitis, D. W., An Asymptotic Analysis of the Induction Phases of Hydrogen-Air Detonations, Combustion Science and Technology, Vol. 52, Nos. 46, 1987, pp. 293323. 119 Grismer, M. J., and Powers, J. M., Numerical Predictions of Oblique Detonation Stability Boundaries, Shock Waves, Vol. 6, No. 3, 1996, pp. 147156.

1229

120 Bourlioux, A., and Majda, A. J., Theoretical and Numerical Structure for Unstable Two-Dimensional Detonations, Combustion and Flame, Vol. 90, Nos. 34, 1992, pp. 211229. 121 Williams, D. N., Bauwens, L., and Oran, E. S., A Numerical Study of the Mechanisms of Self-Reignition in Low-Overdrive Detonations, Shock Waves, Vol. 6, No. 2, 1996, pp. 93110. 122 Helzel, C., LeVeque, R. J., and Warnecke, G., A Modied Fractional Step Method for the Accurate Approximation of Detonation Waves, SIAM Journal on Scientic Computing, Vol. 22, No. 4, 2000, pp. 14891510. 123 Arienti, M., and Shepherd, J. E., A Numerical Study of Detonation Diffraction, Journal of Fluid Mechanics, Vol. 529, 2005, pp. 117146. 124 Short, M., and Stewart, D. S., Cellular Detonation Stability. Part 1. A Normal-Mode Linear Analysis, Journal of Fluid Mechanics, Vol. 368, 1998, pp. 229262. 125 Tegn r, J., and Sj green, B., Numerical Investigation of Deagration e o to Detonation Transitions, Combustion Science and Technology, Vol. 174, No. 8, 2002, pp. 153186. 126 Oran, E. S., Weber, J. W., Stefaniw, E. I., Lefebvre, M. H., and Anderson, J. D., A Numerical Study of a Two-Dimensional H2 O2 Ar Detonation Using a Detailed Chemical Reaction Model, Combustion and Flame, Vol. 113, Nos. 12, 1998, pp. 147163. 127 Liu, J., Oran, E. S., and Kaplan, C. R., Numerical Diffusion in the FCT Algorithm, Revisited, Journal of Computational Physics, Vol. 208, No. 2, 2005, pp. 416434. 128 Kadanoff, L. P., Computational Scenarios, Physics Today, Vol. 57, No. 11, 2004, pp. 10, 11. 129 Buckmaster, J., The Role of Mathematical Modeling in Combustion, Combustion in High Speed Flows, edited by J. Buckmaster, T. L. Jackson, and A. Kumar, Kluwer, Dordrecht, The Netherlands, 1994, pp. 447459. 130 Williams, F. A. The Role of Theory in Combustion Science, Proceedings of the Twenty-Fourth International Symposium on Combustion, The Combustion Inst., Pittsburgh, PA, 1992, pp. 117. 131 Roache, P. J., Quantication of Uncertainty in Computational Fluid Dynamics, Annual Review of Fluid Mechanics, Vol. 29, 1997, pp. 123160. 132 Oberkampf, W. L., and Trucano, T. G., Verication and Validation in Computational Fluid Dynamics, Progress in Aerospace Sciences, Vol. 38, No. 3, 2002, pp. 209272.